ARES dataset for 21cmLSTM

The ARES (Accelerated Reionization Era Simulations; Mirocha et al. 2017; https://github.com/mirochaj/ares) dataset used to train 21cmLSTM, a memory-based global 21 cm signal emulator (see GitHub repository, https://github.com/jdorigojones/21cmLSTM, and original release paper, Dorigo Jones et al. 2024, or DJ+24).

Includes the data (i.e., parameter values) and labels (i.e., brightness temperature values in mK) for the training set, validation set, and test set. There are 8 physical parameters associated with each global 21 cm signal (see Table 1 of DJ+24), and each signal is composed of 449 frequency channels, corresponding to z=5.1-49.9 with redshift resolution of 0.1. The training, validation, and test sets contain 23,896, 2,656, and 1,704 signals, respectively.

For the 21cmGEM dataset that was also used to train 21cmLSTM, see the Zenodo deposit by the 21cmVAE team (Bye et al. 2022): https://doi.org/10.5281/zenodo.5084113.